README.md

metadata

license: apache-2.0
tags:
  - toxicity-detection
  - nlp
  - dataset

ToxicThesis Dataset: VICUNA

Preprocessed dataset for toxicity detection in vicuna outputs.

Contents

This repository contains preprocessed data for training tree-based neural networks:

• preprocessed_data_vicuna.tar.gz - Complete dataset archive

Files in Archive

• preprocessed_data_train.pkl - Training set
• preprocessed_data_val.pkl - Validation set
• preprocessed_data_test.pkl - Test set
• train_dataset_graphs.pt - Training dependency graphs
• val_dataset_graphs.pt - Validation dependency graphs
• test_dataset_graphs.pt - Test dependency graphs
• vocab_stanza_hybrid.pkl - Vocabulary
• Additional metadata files

Usage

# Download and extract
from huggingface_hub import hf_hub_download
import tarfile

file_path = hf_hub_download(
    repo_id="simocorbo/toxicthesis-vicuna-dataset",
    filename="preprocessed_data_vicuna.tar.gz",
    repo_type="dataset"
)

with tarfile.open(file_path, "r:gz") as tar:
    # Use filter='data' for safe extraction (Python 3.14+ requirement)
    tar.extractall(".", filter='data')

Citation

If you use this dataset, please cite the ToxicThesis project.

License

Apache 2.0